• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

It looks like the 'real' /affordable RDNA3 + next gen NV desktop launch won't launch until September. Thoughts?

Soldato
OP
Joined
30 Jun 2019
Posts
7,885
The higher floating point processing power seems legit to me.

There's quite a big performance boost in the Navi33 mobile GPUs also:

7700S.jpg



87 FPS in Cyberpunk 2077 at 1080p, not bad at all.
 
Last edited:
Soldato
OP
Joined
30 Jun 2019
Posts
7,885
So, it's 20% or more boost from RDNA2>RDNA3 at the same CU count.

Edit - It's not a perfect comparison though, because the 6700S is a 28 CU GPU...

So, probably more like 15-20%.

So the RX 7700 XT and RX 7800 XT should be about 20% faster than the last gen.
 
Last edited:
Caporegime
Joined
17 Mar 2012
Posts
48,639
Location
ARC-L1, Stanton System
What i find interesting is the rumoured RX 7950XT and RX 7950XTX.

MLID talked about AMD having fixed some issues with RDNA3 and were preparing for a retape of these GPU's with an increase in performance.

If you look at AMD's fresh RDNA3 APU's the performance of those is mind-blowing for what they are, perhaps soon we will see RDNA3 for how it was intended?

I hope so...
 
Last edited:
Associate
Joined
3 May 2021
Posts
1,237
Location
Italy
What i find interesting is the rumoured RX 7950XT and RX 7950XTX.

MLID talked about AMD having fixed some issues with RDNA3 and were preparing for a retape of these GPU's with an increase in performance.

If you look at AMD's fresh RDNA3 APU's the performance of those is mind-blowing for what they are, perhaps soon we will see RDNA3 for how it was intended?

I hope so...
If that's the case we might see the early rdna3 being phased out very fast
 
Associate
Joined
31 Dec 2011
Posts
862
What i find interesting is the rumoured RX 7950XT and RX 7950XTX.

MLID talked about AMD having fixed some issues with RDNA3 and were preparing for a retape of these GPU's with an increase in performance.

If you look at AMD's fresh RDNA3 APU's the performance of those is mind-blowing for what they are, perhaps soon we will see RDNA3 for how it was intended?

I hope so...

Yeah, its why I am going to hold out for the 7950XT (no rush), I just hope the early adopters are not too ****** if there really is a fault and some silicon changes get these working as intended.
 
Soldato
Joined
9 Nov 2009
Posts
24,982
Location
Planet Earth
It says its RDNA3, and it has a 120 watt TDP, the 6600XT is 160 watts.
Some of you need to stop overhyping Navi 33 like all the overhype with beast mode Polaris cards - I told people where performance it would be like and it was accurate. The ones overhyping it got annoyed and dunked on Polaris.

It's basically a cloned Navi23 with some RDNA3 improvements on 6NM. It will run like an overclocked RX6650XT:

People forget the RX7900 series and 780M IGPs are made on TSMC 5NM class nodes. Navi 33 is made on a denser 7NM class node,so it might not clock as high.The reality is Navi 31,might have reduced the Infinity Cache amount but also it came with a massive jump in memory bandwidth. It had more shaders,higher clockspeeds. It also DOUBLED transistor count over Navi 21.Navi 33 is smaller than Navi 23 on a slightly denser 7NM class node. So the transistor count increase is probably minimal. The memory bandwidth has not increased either.

The RX6650XT has a boost clock of 2.6GHZ:

Even if they manage to get it to run consistently at 3GHZ,that is barely a 15% improvement there.

Add a few percent IPC improvement,and it hits RX6700XT/RX6750XT level performance. So even if were sold as a RX7600XT 16GB,it would be 20% to 25% improvement with more VRAM. Even if it was a 30% improvement,its OK not brilliant.

The move from the RX480 to the RX5600XT/RX5700XT was much bigger.

The issue is if that ROCm leak is true,AMD is trying to use Navi 33 in the RX7700XT. That only means they want to price hike Navi 33 to a higher tier than Navi 23.

So essentially RX6700XT/RX6750XT level performance for the same price in 2023,with a bit extra VRAM. But with all the issues the RX6600 class has as you increase resolution(and also a downgrade to a PCI-E 8X link).

Basically what Nvidia is also doing too.

Oh yeah.... got spot.

21.4 TFlops vs 10.6 TFlops

With a doubling of transistors. Performance in games hasn't doubled.

Don't read into those peak figures. Read commentary on the changes in RDNA3,especially the dual issue nature of the SIMDs,meaning it's not always possible to extract that peak performance figure due to bottlenecks:

I’m guessing RDNA 3’s dual issue mode will have limited impact. It relies heavily on the compiler to find VOPD possibilities, and compilers are frustratingly stupid at seeing very simple optimizations. For example, the FMA test above uses one variable for two of the inputs, which should make it possible for the compiler to meet dual issue constraints. But obviously, the compiler didn’t make it happen. We also tested with clpeak, and see similar behavior there. Even when the compiler is able to emit VOPD instructions, performance will only improve if compute throughput is a bottleneck, rather than memory performance.

I’m guessing RDNA 3’s dual issue mode will have limited impact. It relies heavily on the compiler to find VOPD possibilities, and compilers are frustratingly stupid at seeing very simple optimizations. For example, the FMA test above uses one variable for two of the inputs, which should make it possible for the compiler to meet dual issue constraints. But obviously, the compiler didn’t make it happen. We also tested with clpeak, and see similar behavior there. Even when the compiler is able to emit VOPD instructions, performance will only improve if compute throughput is a bottleneck, rather than memory performance.

Have people forgotten,Ampere also "doubled" FP32 performance too,but in reality it was just more like an accounting methodology change. For instance using the new Nvidia methodology my RTX3060TI has nearly 2.3X the FP32 performance of an RTX2060 Super:

Except in reality it's closer to 40% on average.
 
Last edited:
Associate
Joined
29 Jan 2015
Posts
362
Oh well, at least if the 7600 XT is 32 CUs, it won't be any worse than a 6600 XT:

We know the 7600 is an 8GB card, so it looks likely.

It terms of the performance, the FP32 processing power is about 2x that of the RX 6600 XT:

Basically, they can get a decent boost compared to Navi23, because the RDNA3 architecture is faster.

I wouldn't say we know what 7600 actually is.

We know N33 is 128 bit which mean 8GB or 16GB and it seems logical that N33 will be used in the 7600 but should AMD decide that the 7700 can have 16GB of VRAM and be between 4060Ti and 4070 in price to really push the VRAM narrative then there is space for the 7600 to actually be a cut N32 with 12GB of VRAM and full N33 might end up being used in the 7500XT instead.

OTOH if N33 clocks really well with higher power and better cooling and somehow can manage to hit 6800 tier performance (still very very sceptical that this is the case) then a 16GB 7700XT based on N33 could also make sense but you do lose the option of a 12GB N32 based card unless you want a really wonky lineup.

I expect N33 to be used on 7600 and 7500 with N32 on 7800 and 7700 but there are other possibilities as well. Just depends what price AMD want associated with various names / performance tiers.
 
Caporegime
Joined
17 Mar 2012
Posts
48,639
Location
ARC-L1, Stanton System
Some of you need to stop overhyping Navi 33 like all the overhype with beast mode Polaris cards - I told people where performance it would be like and it was accurate. The ones overhyping it got annoyed and dunked on Polaris.

It's basically a cloned Navi23 with some RDNA3 improvements on 6NM. It will run like an overclocked RX6650XT:

People forget the RX7900 series and 780M IGPs are made on TSMC 5NM class nodes. Navi 33 is made on a denser 7NM class node,so it might not clock as high.The reality is Navi 31,might have reduced the Infinity Cache amount but also it came with a massive jump in memory bandwidth. It had more shaders,higher clockspeeds. It also DOUBLED transistor count over Navi 21.Navi 33 is smaller than Navi 23 on a slightly denser 7NM class node. So the transistor count increase is probably minimal. The memory bandwidth has not increased either.

The RX6650XT has a boost clock of 2.6GHZ:

Even if they manage to get it to run consistently at 3GHZ,that is barely a 15% improvement there.

Add a few percent IPC improvement,and it hits RX6700XT/RX6750XT level performance. So even if were sold as a RX7600XT 16GB,it would be 20% to 25% improvement with more VRAM. Even if it was a 30% improvement,its OK not brilliant.

The move from the RX480 to the RX5600XT/RX5700XT was much bigger.

The issue is if that ROCm leak is true,AMD is trying to use Navi 33 in the RX7700XT. That only means they want to price hike Navi 33 to a higher tier than Navi 23.

So essentially RX6700XT/RX6750XT level performance for the same price in 2023,with a bit extra VRAM. But with all the issues the RX6600 class has as you increase resolution(and also a downgrade to a PCI-E 8X link).

Basically what Nvidia is also doing too.



With a doubling of transistors. Performance in games hasn't doubled.

Don't read into those peak figures. Read commentary on the changes in RDNA3,especially the dual issue nature of the SIMDs,meaning it's not always possible to extract that peak performance figure due to bottlenecks:





Have people forgotten,Ampere also "doubled" FP32 performance too,but in reality it was just more like an accounting methodology change. For instance using the new Nvidia methodology my RTX3060TI has nearly 2.3X the FP32 performance of an RTX2060 Super:

Except in reality it's closer to 40% on average.

CAT we know all this, we are not 5. :) It has dual threads on the cores, like Hyper Threading, similar to how Nvidia did it with Ampere, only they count thiers as physical cores, we soon learned that didn't equate to a doubling of gaming performance.

It will be what it will be, no harm in expressing hope
 
Last edited:
Soldato
Joined
9 Nov 2009
Posts
24,982
Location
Planet Earth
CAT we know all this, we are not 5. :) It has dual threads on the cores, like Hyper Threading, similar to how Nvidia did it with Ampere, only they count thiers as physical cores, we soon learned that didn't equate to a doubling of gaming performance.

It will be what it will be, no harm in expressing hope
Not if the ROCm leak is true and they are upselling Navi 33 and trying to sell it for £400 and then say they have 16GB instead of the RTX4060TI and declare victory.

This is entirely consistent with what they did with the RX7900XT which should have been the RX7800XT.

They should be easily be able to sell an RX7600XT 16GB for around £300. They are selling more expensive to make RX6700XT cards for near that, so their margins will probably go up.

MLID pointed AMD gaming margins are still the same despite the RX6000 series discounts.

With minimal effort they can probably increase sales, still have decent prices and make more money. It depends if the accountants get the last say.
 
Last edited:
Caporegime
Joined
17 Mar 2012
Posts
48,639
Location
ARC-L1, Stanton System
Not if the ROCm leak is true and they are upselling Navi 33 and trying to sell it for £400 and then say they have 16GB instead of the RTX4060TI and declare victory.

This is entirely consistent with what they did with the RX7900XT which should have been the RX7800XT.

They should be easily be able to sell an RX7600XT 16GB for around £300. They are selling more expensive to make RX6700XT cards for near that, so their margins will probably go up.

MLID pointed AMD gaming margins are still the same despite the RX6000 series discounts.

With minimal effort they can probably increase sales, still have decent prices and make more money. It depends if the accountants get the last say.

When did he say that about margins?

AMD published their financials just yesterday, margins are at 44%, that's down 6 points from last quarter, quite a chunk.
 
Soldato
Joined
9 Nov 2009
Posts
24,982
Location
Planet Earth
When did he say that about margins?

AMD published their financials just yesterday, margins are at 44%, that's down 6 points from last quarter, quite a chunk.

Look at the charts. The gaming segment revenue and margins hardly showed a decline. This is despite the selling prices of consoles and RX6000 dGPUs going down compared to the same period last year. So AMD selling those RX6650XT cards for £250 is not at a loss. How could it be when AMD was selling the RX5700 8GB and RX5600XT 6GB for under £300 in 2019? These used a much larger 7NM chip,when 7NM was more expensive and used a 256 bit memory bus,when GDDR6 was new and expensive.

So an RX7600XT(which is probably cheaper to made due to a smaller die),with 16GB of GDDR6 for £300ish(even £320) would be as profitable or more profitable than a £250 RX6650XT or £200 RX6600. So if we end up seeing a £400 Navi 33 product we know what the dGPU cartel is doing.

Most of that margin decline was due to the client CPU sales going down the drain,and probably discounts on Zen3 products. Yet,AMD still seems to be doing better in CPU margins despite undercutting Intel in most areas again.

As the guy from SemiWiki said a few days ago a lot of the costs companies like Nvidia,AMD,Intel,etc have are grossly overestimated on the internet. Remember end user sales of PC parts are probably at a much higher margin compared to companies like Dell, who will drive down costs,and are not swayed by RGB and Waifu Edition cards.

RX7600XT 16GB ? Seems high was thinking 12gb ? Still more what 4060ti 8gb offers still can say they increased from 8gb last gen

Navi 33 uses a 128 bit memory controller like Navi 23. It is basically an RDNA3 shrink of Navi 23 onto a modified 7NM class process node.

IIRC,there was noise about GDDR6/GDDR6X(or was it GDDR7?) offering mixed memory configurations but no one has implemented it.

Which is a shame! :(
 
Last edited:
Associate
Joined
13 Sep 2010
Posts
2,093
What i find interesting is the rumoured RX 7950XT and RX 7950XTX.

MLID talked about AMD having fixed some issues with RDNA3 and were preparing for a retape of these GPU's with an increase in performance.

If you look at AMD's fresh RDNA3 APU's the performance of those is mind-blowing for what they are, perhaps soon we will see RDNA3 for how it was intended?

I hope so...
"Super-RDNA3" :D
 
Soldato
OP
Joined
30 Jun 2019
Posts
7,885
I wouldn't say we know what 7600 actually is.

We know N33 is 128 bit which mean 8GB or 16GB and it seems logical that N33 will be used in the 7600
I doubt they would bother with more than 8GB of VRAM, it seems overkill on a N33 GPU. It's a 1080p class GPU (according to the AMD slides for N33 mobile).

16GB will be reserved for N32 graphics cards.

20GB and above for N31.
 
Last edited:
Soldato
OP
Joined
30 Jun 2019
Posts
7,885
I think you would have seen 16GB of VRAM on the N33 mobile GPUs if that was the plan.

12/16GB is a big selling point for N22 and N32 cards.

I bet the 7700 is gonna be reasonably priced tbf.
 
Last edited:
Back
Top Bottom